Interesting IETF drafts
We’re watching some interesting drafts lately:
- LEDBAT: Low extra delay background transport. Makes you wonder why more people haven’t thought about this problem (of course our hostility towards TCP is well known here)
- ALTO: Bringing cost awareness to the network and helping bridge application and network domains. This has been a sore spot for a long time that we need to do a more detailed overview on.
- More in the same vein: P2P Edge Caching and its effect on network economics.
74th IETF next week
The 74th IETF starts March 22nd (next week) in lovely San Francisco, CA.
Meeting agenda is here. We’ll do our best to summarize the minutes and meetings that will be taking place although we won’t be there in person.
Feature Spotlight: IOS Enhanced Object Tracking
The Feature Spotlight will be a regular series here at TCPMSS
that takes a deeper dive into a feature of a vendor’s equipment.
TCPMSS is just highlighting these features, and can’t speak as
to their completeness, reliability, or universality among a
particular vendor’s equipment.
Enhanced Object Tracking has been a feature in IOS for quite a while but in combination with its sibling, Embedded Event Manager, EOT can be used to do some pretty amazing things. We’ll talk about EEM in a future post in this series.
Historically, HSRP and other reliability protocols only tracked interface state. EOT extends this to several other useful items and introduces an abstraction within IOS to permit use of different objects to trigger a variety of events on the router for First Hop Reliability Protocols (FHRP). Reportedly, this also provides a significant performance gain on the control plane, preventing multiple protocols from needing to track the same state change on a device.
The following are examples of items that can be tracked using Enhanced Object Tracking:
- IP SLA (if an item is over threshold or down, this can be used as a state change within EOT)
- IP Route Metrics: If a metric goes over the threshold, state notification changes
- Reachability: If a route’s next hop isn’t reachable, state change
- Interface state: This one was already there.
You can get really crazy, too, with how all these items interact. You can use arbitrary weights and percentages, or boolean operators, which brings more programmability to the router. The IOS manual on this feature has a good example (at least of threshold weighting):
track 1 interface serial2/0 line-protocol
track 2 interface serial2/1 line-protocol
track 3 interface serial2/2 line-protocol
exit
track 100 list threshold weight
object 1 weight 20
object 2 weight 20
object 3 weight 20
threshold weight down 0 up 40
This stuff is very useful on any GLBP/HSRP/VRRP device.
NANOG45 Summary
NANOG45 in the Dominican Republic ended on January 28. Presentations are here.
Sunday
Dino Farinacci + Dave Meyer: LISP+ALT
Sunday was a tutorial day. Dave Meyer and Dino Farinacci gave a summary of LISP, which is an abstraction layer for IP routing that separates the “locator” for an address from the address itself. The argument is to split host and network/router stacks, based on the problem statement elucidated by Vince Fuller an APRICOT plenary. The basic summary of the problem is that DRAM isn’t the problem, fast TCAM growth is. We talked about TCAM here in our router architecture primer (part I).
Dino & Dave’s approach has come a long way (and the fact that there is working code and a valid implementation that has seen some proving is great). Now that their LISP+ALT effort has some ties into the “legacy” world things are great, but some work on automating the process is needed. There’s still a challenge of seeing this approach widely adopted by other vendors, and precious little research on what i’d call “legacy effects”, meaning what happens from a scale perspective when LISP+ALT isn’t deployed everywhere. I think maybe the problem is already solved; if Dino can get adoption by the inner circle of state depicted in his and Dave’s presentation, the problem won’t go away but it will buy us some time. I wasn’t able to watch the video of the presentation (woefully tied down away from the DR this NANOG), so it’s hard to say whether there were definitive statements from providers or not. But having these statements would be a useful exercise.
Steve Gibbard et al, Network Operations Practices
I was impressed by the presentation. It was more of a tutorial (which is appropriate since this is a tutorial day). I wish we could see more details of how the network operations efforts at different providers play out, and a little more modeling of the positive or negative benefits of instituting process, but overall he covers basically eveyrhting you need to know about “Network Operations for Dummies”. Great job.
Freedman & Jaeggli/Toonk presentations
I won’t cover these in any exhaustive detail. Avi gave a good scaling presentation on BGP’s mechanisms. The BGP hijacking tools presentation was also good. I’m surprised we haven’t seen a better database on hijackings short of the reports that are automatically sent out to the NANOG mailing list.
Monday (Day 2)
Meyer’s End of the World
Dave Meyer gave a great talk about the scaling issues facing the Internet. He broke it down into data and control plane scaling, and then further into “What should have happened [or was expected to happen]” vs “what actually happened”. I like how he didn’t turn it into an advertising session for LISP. He highlighted the need for research into CGN (Carrier Grade NAT, see Randy Bush’s presentation for more detail) and alternate approaches for Internet-scale data & control planes that can actually be categorized as “deployable”. Question still remains, who is going to perform this research. Action is needed.
Tom Scholl/ATT Labs BFD
Tom did a really great presentation here on BFD, not only introducing the topic, but talking about contrasting options for subsecond detection and its interaction with other control plane protocols. The conclusion is that BFD is a great tool to use in your arsenal for network recovery. There wasn’t much discussion about the failure intervals appropriate for services used on AT&T’s network, something I desperately wanted to know more about, and Tom kind of genericised the vendor support. BFD certainly isn’t deployed on all platforms by vendors equally.
Renesys Instability Scoring
Jim Cowie with Renesys presented some extremely good data and analysis on individual providers and an overview of scoring for instability. Basically, moving beyond simple prefix flapping and into a more generic architecture for how to analyze network stability. Absolutely flabbergasting that Cogent was ahead of Sprint! Now we need to move on to practical suggestions about how providers can improve their scores relative to the pack (although you could infer how to do that from the deck).
I broke out of the session at this point, but you definitely need to check out Andy Davidson’s presentation on ASN4 (for more information on 4-byte ASNs, Greg Hankins did a presentation on Tuesday that covered it) and Phil Roberts’ presentation on IPv4/IPv6 issues if you’re concerned about Internet scalability.
Tuesday (Day 3)
Martin Levy, HE - IPv6
Martin gave a preso on IPv6 deployment at HE, and the traffic levels they’re seeing. Worth checking out.
Scudder’s BGP Monitoring Draft
John Scudder used to work at Cisco if I recall correctly; he presented with another colleague from Juniper Networks the BGP Monitoring Protocol. Now BMP, or its antecedents, have been around for a while; it always floored me that Cisco never made any choice to deploy it despite the fact that it was more than worthwhile. I think this is one of many nails in the coffin that affected Cisco’s SP futures. I’m not surprised that after Scudder left Cisco that he went on and did this work at Juniper.
Submarine Cable Panel
This was a really great session that opened up the details of the submarine cable market; it follows a session at an earlier NANOG with more details about submarine cables. For a lot of operators this stuff is a black art, and I cannot express how happy I am that details of FEC, amplification, and market drivers in submarine cables saw the light of day. Probably put a lot of people to sleep, but super valuable for those thinking about investing (or getting involved in investment consortiums). There are plenty of mid-size SPs that could afford this level of investment, particularly if they worked together.
Popescu, Hepner, and Brown @ Renesys: De-Peering
A detailed exposition of the Cogent-Sprint peering debacle earlier this year. Great views on why this is bad. There’s responsibility in being a provider in the default-free zone, and the effects need to be well understood before you depeer someone. Exit strategies need to be understood.
Brandon Ross Lightning Talk: XIOCOM
These guys came up on my radar a year or two ago. They’re building wireless networks in Emerging Markets. Good information if you want to understand some of the challenges in these deployments.
Kotikapaludi Sriram, et al.: BGP Anomaly Detection and Robustness
KS and partners presented a good, good view of some of the BGP anomaly detection algorithms out there today (including prefix hijack detection). So in the context of the earlier Jaeggli/Toonk presentations on BGP hijacking, there might be a workable solution here that could use some more attention from the operator community.
Hope you liked our summary! Stay tuned for more this week.
NANOG45 Summary on the way…
We’ll have our summary of NANOG45 done on Monday for your reading pleasure. And NANOG46 will be in Philadelphia! Love that town.
The NOC Project
The NOC Project is a new open source network management system that seems to finally have some clue and scalability, with a smart web interface written in Django which should make it fairly suitable for redeployment. The feature list is long but abstracted, and it seems to support the things SPs commonly need: RPSL, AS-SET objects, peering management, IP address management and DNS provisioning, and fault management (including event correlation!) It’s only at a very early release state but looks fairly interesting even at this phase.
Administrivia for January 2009
We’ve upgraded our host and changed the theme around temporarily while our “final” theme is being worked on. Thanks for your patience and let us know if anything doesn’t work.
